1. Lecture 5. Niching and Speciation (2) 4 학습목표 진화로 얻어진 해의 다양성을 확보하기 위한 대표 적인 방법에 대해 이해한다.

2. Outline  Review of the last lecture  A motivating example of niching: co-evolution in classification tasks  Why niching  Different niching techniques  Sharing and crowding  Relationship between niching and speciation  Summary

3. Different Niching Techniques  Can be divided roughly into two major categories  Sharing, also known as fitness sharing  Crowding  Other niching methods include sequential niching and parallel hillclimbing

4. Fitness Sharing: Introduction  Fitness sharing transforms the raw fitness of an individual into shared fitness  It assumes that there is only limited and fixed “resource” available at each niche. Individuals in a niche must share them  Sharing is best explained from a multimodal function optimization perspective individual raw fitness How can we locate multiple peaks in one evolutionary process?

5. Fitness Sharing: Implementation  Define a sharing radius  share : Anything within this radius will be regarded to be similar to the individual and thus needs to share fitness  Define a similarity measure, i.e., distance: The shorter the distance between two individuals, the more similar they are  Define a sharing function  Define shared fitness an individual  share The individual in the center needs to share fitness with all other in the circle sh(d) = 1 – (d /  share ) if d <  share  0 otherwise distance a scaling constant f share (i) = f raw (i) /  j=1 sh(d ij )  where  is the population size

6. Fitness Sharing: Extensions  Sharing can be done at genotypic or phenotypic level  Genotypic: Hamming distance  Phenotypic: Euclidean distance (Overlap in test case covering in classification)  The key issue is how to define the “distance” measure  Sharing radius  share can be difficult to set, the same, fixed: It should be sufficiently small in order to discriminate between two neighboring peaks  Population size should be sufficiently large to locate all peaks  Population may not be able to converge to exact optima  Population may not be stable, i.e., may lose peaks located  Calculate shared fitness needs time  Fitness sharing often needs raw fitness scaling. f share (i) = f raw (i) /  j=1 sh(d ij )   scaling factor Why?

7. Why Fitness Scaling  Let f i ’ = f share (i), f i = f raw (i), m i =  j=1 sh(d ij )  Then fitness sharing is f i ’ = f i /m i  Fitness sharing with scaling is f i ’ = f i  / m i,  >= 1  top-down view of a 2-d space side-view of the 2-d space raw fitness shared fitness without scaling with scaling using sufficiently large 

8. A Dilemma  With low scaling factor: individuals won’t go to the real optimum because it’s not attractive  With high scaling factor: We may not be able to find all peaks, because a high scaling factor creates “super individuals”, even a very soft selection scheme won’t help  A possible solution: Anneal the factor  1.Start the evolution with a small , e.g.,  = 1, in order to explore and locate the peak regions 2Then increase  gradually to attract individuals to the optima

9. Implicit Fitness Sharing (1)  The idea comes from an immune system: antibodies which best match an invading antigen receive the payoff for that antigen  Similar situation occurs in games: a strategy receives payoff when it achieves the best score against a test case  Implicit fitness sharing is most often used in learning. While (explicit) fitness sharing is done through individuals, implicit fitness sharing is test data based!  The algorithm for calculating fitness: For each data point i to be matched, do the following C times 1. Select a sample of  individuals from the population 2. Find the individual in the sample that achieves the highest score against the data point i 3. This best individual receives the payoff. In the case of a tie, payoff is shared equally

10. Implicit Fitness Sharing (2)  It has been shown that implicit and explicit fitness sharing have the same theoretical basis.  here plays the role of  share in (explicit) fitness sharing  Larger C  better result but more time-consuming  Comparison between implicit and explicit sharing: they are better under different circumstances  Implicit fitness sharing covers optima more comprehensively, even when those optima have small basin of attraction, when the population is large enough for a species to form at each optimum  (Explicit) fitness sharing can find the optima with larger basins of attraction and ignore the peaks with narrow bases, when the population is not large enough to cover all optima

11. Niching vs. Speciation  Although some people distinguish between the two, we will treat them as the same thing  If there is any difference:  niching is concerned more with locating peaks (basins of attraction), while speciation is more focused on actually converging to optima

12. Summary  Co-evolution is closely related to fitness sharing although fitness sharing was first motivated by multimodal function optimization  Niching and speciation are useful  There are different niching methods  All niching methods involve fitness evaluation. However, they do interact with selection and crossover  References  T. Back, O.B. Fogel and Z. Michalewicz, Handbook of Evolutionary Computation, IOP Pub. Ltd & Oxford Univ. Press, 1997. Section C6.1 and C6.2 (only 9 pages)  P. Darwen and X. Yao, “A dilemma for fitness sharing with a scaling function,” Proc. Of IEEE ICEC, 1995, pp. 166~171  P. Darwen and X. Yao, “Every niching method has its niche: fitness sharing and implicit sharing compared,” LNCS, vol. 1141, pp. 398~407, 1996.